The comparision doesn’t work. Because the AI is replacing the pencil or other drawing tool. And we aren’t saying pencil companies are selling you Mario pics because you can draw a Mario picture with a pencil either. Just because the process of how the drawing is made differs, doesn’t change the concept behind it.
An AI tool that advertises Mario pcitures would break copyright/trademark laws and hear from Nintendo quickly.
I don’t think how you interact with a tool matters. Typing what you want, drawing it yourself, or clicking through options is all the same. There are even other programs that allow you to draw by typing. They are way more difficult but again, I don’t think the difficulty matters.
There are other tools that allow you to recreate copyrighted material fairly easily. Character creators being on the top of the list. Games like Sims are well known for having tons of Sims that are characters from copyrighted IP. Everyone can recreate Barbie or any Disney Princess in the Sims. Heck, you can even download pre made characters on the official mod site. Yet we aren’t calling out the Sims for selling these characters. Because it doesn’t make sense.
I don’t buy the pencil comparison. If I have a painting in my basement that has a distinctive style, but has never been digitized and trained upon, I’d wager you wouldn’t be able to recreate neither that image nor it’s style. What gives? Because AI is not a pencil but more like a data mixer you throw complete works in into and it spews out colllages. Maybe collages of very finely shredded pieces, to the point you could even tell, but pieces of original works nontheless. If you put any non-free works in it, they definitely contaminate the output, and so the act of putting them in in the first place should be a copyright violation in itself. The same as if I were to show you the image in question and you decided to recreate it, I can sue you and I will win.
That is a fundamental misunderstanding of how AI works. It does not shred the art and recreate things with the pieces. It doesn’t even store the art in the algorithm. One of the biggest methods right now is basically taking an image of purely random pixels. You show it a piece of art with a whole lot of tags attached. It then semi-randomly changes pixel colors until it matches the training image. That set of instructions is associated with the tags, and the two are combined into a series of tiny weights that the randomizer uses. Then the next image modifies the weights. Then the next, then the next. It’s all just teeny tiny modifications to random number generation. Even if you trained an AI on only a single image, it would be almost impossible for it to produce it again perfectly because each generation starts with a truly (as truly as a computer can get, an unweighted) random image of pixels. Even if you force fed it the same starting image of noise that it trained on, it is still only weighting random numbers and still probably won’t create the original art, though it may be more or less undistinguishable at a glance.
AI is just another tool. Like many digital art tools before it, it has been maligned from the start. But the truth is what it produces is the issue, not how. Stealing others’ art by manually reproducing it or using AI is just as bad. Using art you’re familiar with to inspire your own creation, or using an AI trained on known art to make your own creation, should be fine.
As a side note because it wasn’t too clear from your writing, but the weights are only tweaked a tiny tiny bit by each training image. Unless the trainer sees the same image a shitload of times (Mona Lisa, that one stock photo used to show off phone cases, etc) then the image can’t be recreated by the AI at all. Elements of the image that are shared with lots of other images (shading style, poses, Mario’s general character design, etc) could, but you’re never getting that one original image or even any particular identifiable element from it out of the AI. The AI learns concepts and how they interact because the amount of influence it takes from each individual image and its caption is so incredibly tiny but it trains on hundreds of millions of images and captions. The goal of the AI image generation is to be able to create vast variety of images directed by prompts, and generating lots of images which directly resemble anything in the training set is undesirable, and in the field it’s called over-fitting.
Anyways, the end result is that AI isn’t photo-bashing, it’s more like concept-bashing. And lots of methods exist now to better control the outputs, from ControlNet, to fine-tuning on a smaller set of images, to Dalle-3 which can follow complex natural language prompts better than older methods.
Regardless, lots of people find that training generative AI using a mass of otherwise copyrighted data (images, fan fiction, news articles, ebooks, what have you) without prior consent just really icky.
